Enterprises are turning to tokenization to address data security challenges that are slowing the deployment of AI agents.
Tokenization could help enterprises accelerate AI agent deployments by protecting sensitive information without preventing models from using it, according to Diego Loureda, Protegrity’s vice president of customer success and presales.
In a recent interview with theCUBE, Loureda warned that once sensitive data enters an AI model, vector store or an agent’s memory, it can propagate through connected systems and become difficult to remove. As companies move more agents into production, data protection must therefore account for how information is used and shared across AI workflows.
Protegrity’s technology replaces sensitive values with substitutes, rather than relying on conventional encryption that can restrict access to data. Loureda said the method allows protected information to remain available for analytics and AI applications. He also described tokenization as quantum-ready because it uses value replacement in memory instead of a mathematical algorithm.
The need for more scalable protection is underscored by research commissioned by Protegrity from Enterprise Management Associates. According to Loureda, 83% of leaders surveyed reported delays in their projects, while 82% said they were deploying AI agents with reduced capabilities after removing tools and data to address security concerns.
Loureda said companies should shift from approving individual AI projects to approving reusable governance and security patterns that can support multiple deployments. He argued that tokenization can help security teams maintain control over sensitive data while allowing AI agents to access the information they need, making security an enabler of enterprise AI adoption rather than a final approval hurdle.
